List your company
ResearchFormerly WizWand

Speech Question Answering on Llama Questions

64.7Speech QA ScoreGLM-4-Voice
19.25231.05142.8554.649Aug 10, 2025Sep 29, 2025Nov 19, 2025Jan 9, 2026Feb 28, 2026Apr 20, 2026Jun 10, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GLM-4-Voice2025.0864.7
TurnGuideloss ratio=2:12025.0864
Moshi2025.0862.3
Scaling InterleaveModel=9B, Train Params=9B, Data=600B tok2026.0650.7
Scaling InterleaveModel=9B, Train Params=9B, Data=200B tok2026.0644
Ours (8B)Model=8B, Train Params=~150M, Data=2.5k hrs2026.0639.3
Scaling InterleaveModel=9B, Train Params=9B, Data=100B tok2026.0637
Ours (4B)Model=4B, Train Params=~100M, Data=2.5k hrs2026.0630.7
Spectron2025.0821.9
SpeechGPT2025.0821.6
MoshiModel=7B, Train Params=7B, Data=7M hrs2026.0621